DogFoodHackathon2k26

I'm writing a technical write-up for the DOGFOOD 2026 Write-Up Side Quest.
Judging is on insight, not polish or follower count — a small honest post
beats a viral thread that says nothing. The brief asks for one of: an
architecture decision that shaped everything, a Docker gotcha that cost
hours, a scoring algorithm I almost shipped before realizing it was
broken, or a part of the spec that looked simple until I actually built it.

Before writing anything, investigate the actual project history, not just
the current state:

• Look at git log / commit messages for signs of reverted approaches, rewrites, or "fix:" commits that suggest something broke and got patched
• Compare early commits of JUDGING.md, DATA-MODEL.md, or the scoring/auth code against the final version — what changed and why
• Check for any commented-out code, TODOs, or abandoned files that hint at a path we started and dropped
• Look specifically at the normalization logic, the peer-isolation checks, and anything touching fixtures.json's awkward cases (the duplicate submission, the zero-variance judge, the uneven review counts) — these are the likely sites of a real "it looked simple until I built it" moment

Report back 2-3 candidate moments with the actual evidence (commit hashes,
file diffs, specific bug behavior) rather than a generic narrative. For
each, tell me:

1. What we initially assumed or tried
2. What broke it / why it was wrong
3. What we shipped instead and why that's defensible

Don't write the full post yet — show me the candidates first so I can
pick the one that's most honest and specific, then I'll ask you to draft it.